Then your average value estimation would also suffer from the time lag distortion I've described.
As I said repeatedly my model was simplified and the effects would be blunted in the real world, but it would be there. And no I never said the time lag would sudden increase in death cases at the end-stage. Rather it would lead to a higher daily death/daily discharged ratio toward the end of outbreak than it would otherwise be. I didn't say anything about the absolute number and I didn't say anything about the increase being 'sudden'.
The data outside Hubei isn't very helpful because the number of death outside Hubei is just too small. The time lag effect would be easily obscured. Let's wait for the Wuhan data. We're not that impatient, ain't we? Of course, even if we have the Wuhan data we couldn't really attribute its behaviors simply to the time lag or any other factor without doing a lot more research, which I have no time for. And even if the data turns out doesn't behave as I have predicated it doesn't mean the distortion effects ain't there. There might just be overwhelming counteracting factors. Sucks but working with real world data is always messy.
In any case, the increase in the proportion of serious case is probably not due to 'the way they define it so as to provide better care to the patients at the later stage of the epidemic when ample resources are available'. We have been on the 7th edition of NHC's COVID-19 diagnosis and treatment guideline for a while now, and the criteria for serious cases are set by the national guideline.
No, more likely the increase is a result of
"Using available preliminary data, the median time from onset to clinical recovery for mild cases is approximately 2 weeks and is 3-6 weeks for patients with severe or critical disease. "
Mild cases recovers much faster than serious cases. Since the number of new cases keeps decreasing, the proportion of serious cases keep increasing. As the proportion of serious cases increases,
caeteris paribus, the daily death/daily recovery ratio goes up. This is how the time lag distortion works. Really I feel I've made a good prima facie case why there would be a time lag distortion (whose existence seems to me to be rather obvious). Shouldn't you bear the burden to give reasons why you think there isn't one?